Accurate scores · smooth detection · UX improvements

AI Detector upgraded from v17b to v18 — Korean vocal false-positive rate reduced from 70% to 0%
Anti-AI processing now produces a smooth score drop (previously swung between 0% and 99%)
Score display capped at 5–99% — no detector can be 100% certain; we reflect that honestly
Detection speed improved 16× (40 s → 2.5 s per track)
Before/After comparison now shows an Anti-AI effect badge (e.g. ↓ −15 pt)
Preset label shown dynamically in the After panel (e.g. 'After · Hip-Hop')
Fixed a cache bug where re-uploading a different track could download the previous master
Eliminated digital click at the start and end of mastered files (stronger fade)
Temp files cleaned up automatically — your audio never persists on disk after processing
Added a re-select file button — switch tracks without a full page reload
Landing page copy made more honest — changed '96%→14%' single-case claim to '15–50% range'
Changelog page launched (/changelog) — cumulative update history

Changelog